What the AI system is, and what it does
Name
Oregon Health Plan (OHP) Work or Activity Rules Chatbot
What it does
The chatbot helps people understand federal Medicaid work or activity rules for keeping OHP coverage. These rules take effect Jan. 1, 2027.
You can ask the chatbot questions about:
- Activities that may count toward meeting the work or activity rules.
- Reasons the rules may not apply to you
- How to report your hours or activities.
The chatbot provides information only. It does not decide if you qualify for OHP or whether you can keep your health coverage.
Who can use it
The chatbot is designed for OHP members and people applying for OHP. Anyone can use it.
The chatbot is available only in English right now.
You can get help in other languages. There
are options for getting help in languages other than English.
How it is accessed
You can use the chatbot when signed into your ONE Online account. You can also use it without signing in.
You can access the chatbot through:
- Your ONE Online account.
- The Oregon ONE Mobile app.
- The ONE Customer Service Center, by phone.
Go to
benefits.oregon.gov for help with setting up your ONE Online account, downloading the mobile app, or to find contact information.
Help with questions is also available by phone or in person:
System owner
ODHS is responsible for the chatbot. Deloitte Consulting LLP built and supports the chatbot under a contract with the Oregon Eligibility Partnership (OEP) within ODHS.
Data used
The chatbot is designed to use only public information or information approved for limited use. It is not designed to use private health information or other highly sensitive information.
The chatbot is also designed to detect and remove personal or private health information you may enter by mistake before it is sent to the AI model.
The state classifies information as:
- Level 1 Published
- Level 2 Limited
- Level 3 Restricted
- Level 4 Critical
The chatbot is not designed to include Level 3 or Level 4 information. The state does not send the AI model:
- Personally Identifiable Information (PII).
- Information protected under the Health Insurance Portability and Accountability Act (HIPAA) or the Family Educational Rights and Privacy Act (FERPA).
- Criminal Justice Information Services (CJIS)
- Federal Tax Information (FTI).
Even with these protections, do not enter personal or private information if it's not needed to ask your question.
Data residency
Information from the chatbot is stored, processed and accessed only in the United States. This includes questions and answers, system logs, and backups.
AI model used and model card
The chatbot uses AI models from Anthropic including Claude Sonnet. You can learn more about these models through Anthropic’s
model system cards.
No model training
Your questions and the chatbot's answers are not used to train or update the AI models. The chatbot does not learn from your individual conversations.
Major risks identified for mitigation
There are some risks when you use the chatbot:
- The chatbot may give a wrong answer.
- You may enter personal or private health information by mistake.
- The chatbot may sometimes be slow.
We test and monitor the chatbot to reduce these and other risks.
Testing snapshot
The chatbot was tested before making it available to the public. It continues to be tested while it is in use.
Testing includes:
-
Accuracy and quality: Experts compare answers with approved information.
-
Fairness: The chatbot is tested to check whether it gives consistent answers to people in different situations.
-
Privacy: It is tested to confirm it detects and removes personal and private health information.
-
Security: Protections against attacks and misuse are tested.
-
System use: The chatbot's use of computing resources is tested.
The chatbot will be tested when there are major changes, including:
- Major policy changes.
- Changes to the AI models.
- Changes to the instructions given to the AI model.
- Other significant system changes.
- An increase in reported problems.
User feedback
Your feedback helps us find problems and improve the chatbot.
After each answer from the chatbot, you can select a thumbs-up or thumbs-down. You can also provide more detailed feedback.
A team regularly reviews feedback to find and fix problems to improve the chatbot.
Monitoring
The chatbot is monitored for problems. This includes:
- Security threats and attacks.
- Whether personal or private health information is kept out of the system
- Use of system resources.
The chatbot can be turned off at any time.